Recommender Systems with AI: From Collaborative Filtering to Deep Learning

A premium, complete and hands-on course on modern recommender systems, updated for 2026. You will start from why recommendation is one of the highest-leverage applications of machine learning, then build up every major family of models with real Python code. You will master content-based filtering with TF-IDF and embeddings, memory-based collaborative filtering (user-based and item-based neighborhoods), and matrix factorization with SVD, ALS and Bayesian Personalized Ranking using the implicit and LightFM libraries. You will learn the crucial difference between explicit and implicit feedback and why it changes both your model and your loss. From there you will build deep learning recommenders — neural collaborative filtering, embeddings, and the two-tower retrieval architecture — and sequential and session-based models with self-attention (SASRec, BERT4Rec). You will design the two-stage candidate-generation-plus-ranking architecture that powers recommendation at scale, serve it with approximate nearest neighbor search, and solve the cold-start problem. You will evaluate systems correctly with precision@k, recall@k, MAP and NDCG, understand why offline and online metrics diverge, and run trustworthy A/B tests. The final modules cover how large language models reshape recommendation in 2026, how to build for diversity, serendipity and fairness, and how to stay compliant with the GDPR when you profile user behavior. Includes a comprehensive final assessment. This deepened edition adds factorization machines, graph neural networks (LightGCN, PinSage), multi-armed and contextual bandits with off-policy evaluation, and a dedicated lesson on the EU Digital Services Act's recommender transparency duties - around 30 in-depth lessons of university-level, practice-first material.